OpenAI released GPT-6 Sol and GPT-6 Luna on September 22, extending the GPT-6 family below Astra. The practical change is a lower cost for recurring agent work, with two models aimed at different workloads. The API identifiers are gpt-6-sol and gpt-6-luna. OpenAI’s announcement
The choice is worth testing at the task level. A cheaper request helps only if the resulting work meets the same acceptance criteria without enough retries or human correction to erase the saving.
Shared capacity, different prices
The model pages list a 1,050,000-token context window and 128,000-token maximum output for both models. Sol targets complex coding and agent workflows; Luna targets focused tasks at high volume. Both accept text and images and produce text. Sol specifications, Luna specifications
Standard prices per million text tokens are:
| Token category | GPT-6 Sol | GPT-6 Luna |
|---|---|---|
| Input | $2 | $0.10 |
| Cached input | $0.20 | $0.01 |
| Cache writes | $2.50 | $0.125 |
| Output | $10 | $0.50 |
Those are base rates. For prompts exceeding 272,000 input tokens, both model pages specify doubled input and cache rates and 1.5 times the output rate for the whole request. Sol pricing, Luna pricing
For illustration, 100,000 uncached input tokens and 10,000 output tokens would cost $0.30 on Sol or $0.015 on Luna at those standard rates. That calculation excludes tools, cache writes and any other processing premium. It compares an identical token budget, not equal task performance.
A large context window also does not establish that a model will retrieve every relevant detail reliably. Test the specific documents or code paths the application needs it to use, especially when a missed detail can survive into a plausible-looking answer.
Check the API path before switching
Both models support reasoning effort from none through max, with medium as the default. The model pages direct developers to the Responses API for built-in tools and function calling. On Chat Completions, function calling requires reasoning effort to be none. Sol API guidance, Luna API guidance
That restriction matters to an application which assumes it can change only the model name while retaining its existing reasoning and tool-call configuration. Exercise an actual tool round trip, including the follow-up response, before broadening a rollout.
OpenAI also says GPT-6 caching can retain reusable context when developers change reasoning effort or enable and disable tools. The announcement describes diagnostics and explicit cache breakpoints for investigating reuse. Caching changes
Measure cache behavior over a complete session. A single warm request will not show whether a frequently changing prefix undermines reuse during normal work.
Where the models are available
At launch, Sol and Luna are rolling out to ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users. Free and Go users can access Luna in the desktop app. OpenAI says the models are not yet available in Chat and that rollout may take the day. Launch availability
We have not independently benchmarked either model. For a first comparison, choose a recurring task with an observable finish condition and run it at a fixed effort setting. Record failures and required edits along with token cost. Keep the current model available until the new one meets the same standard on representative work.
Our Grok 4.7 migration article covers the same accounting problem from another provider’s release: the useful unit is the completed, checked job.
September 25 update: rerun image evaluations
OpenAI reported an image-encoding defect that had degraded Sol and Luna’s image understanding and says it fixed the issue on September 25. The fix covers the API and Codex, including computer use. OpenAI recommends rerunning image evaluations and retrying affected workflows. OpenAI API changelog
If an evaluation used screenshots or other images before the fix, retain that result as a dated baseline and rerun the same inputs. Compare task completion and visual errors under the same settings before changing a deployment decision. This update concerns image processing; it does not revise the launch prices or context specifications above.